The Impact of Income-Related Medicare Part B Premiums on Labor Supply

Abstract

The 2003 Medicare Modernization Act introduced income-related premiums on Medicare coverage for professional services (Part B) for the first time. Beginning in 2007, higher-income households were required to pay higher premiums for Part B coverage, which raises the price of Medicare relative to employer-sponsored health insurance for these households. The authors exploit this exogenous change in Medicare policy to examine the impact of Part B premiums on the labor supply decisions of older adults. They find that higher Medicare premiums delay retirement. Findings have important implications for Medicare policy and labor markets.

Keywords

Medicare Part B insurance premiums retirement instrumental variable means testing

Historically, Medicare Part B—the part of Medicare that covers professional services, primarily physicians’ services—has been available to all eligible persons on the same terms. The Balanced Budget Act of 1997 set premiums at 25% of Part B program cost. Given increased program outlays, Part B premiums had been increasing over time but at the same rate for all enrollees. The philosophy of not varying premium contributions by income changed in 2003 with enactment of the Medicare Prescription Drug, Improvement, and Modernization Act (MMA), which introduced income-related premiums for Part B coverage for the first time. The Deficit Reduction Act of 2005 accelerated the implementation of these income-related premiums. Beginning in 2007, higher-income households were required to pay higher premiums for Part B coverage equal to 35, 50, 65, or 80% of total costs. With this change in policy affecting only a small proportion (4%) of enrollees,¹ this was a modest step toward addressing the long-term financing problem facing Medicare.

Recent discussions about health care financing and increasing costs have focused on further means testing as a potential strategy. The Affordable Care Act (ACA) of 2010 included provisions that will increase the number of Medicare beneficiaries facing higher income-related premiums to 14% by 2019. Specifically, the ACA froze the Part B income thresholds at 2010 levels through 2019 (effective 2011). In addition, it introduced income-related premiums for Medicare Part D (prescription drug coverage), a provision estimated to generate $25 billion in program savings from 2010 to 2019 (Kaiser Family Foundation Issue Brief No. 8126). Means testing could increase revenue without substantially restricting access to personal health care services. Opponents argue that means testing would lead to high-income retirees dropping coverage, which would exacerbate the adverse selection problem since higher-income individuals tend to be healthier (Kaplan 2006). Moreover, if the revenue generated from means testing is to have any meaningful impact on government net expenditures, the income thresholds for higher premiums would have to be fairly low. This would mean that middle-income households could face increased premiums, a politically undesirable option. Alternatively, the income threshold could be high but with a corresponding increased burden on few high-income households, which could further intensify the adverse selection problem.

The shift toward means testing, although presently affecting few beneficiaries, raises the price of Medicare relative to private health insurance, particularly employer-based insurance. This relative price change could influence the retirement decisions of Medicare age-eligible persons. In this study, we use data from the Health and Retirement Study (HRS) to examine the impact of income-related Part B premiums on decisions about labor force participation (LFP) among persons aged 55 to 80. Our study is the first to examine the impact of means-tested coverage within the Medicare program on the labor supply decisions of older workers. Older workers are an important group to focus on since they may be less tied to the labor market than other groups that have been studied extensively in the context of means testing, such as single mothers affected by Medicaid expansions. Further, our study examines the impact of an increase in the price of public insurance in contrast to most prior work that evaluated increases in private insurance premiums. Given these differences, findings from existing studies may not generalize to this study’s target population.

Effect of Insurance Premiums on Labor Supply

Whether or not to enroll in Medicare Part B, as opposed to Part A (coverage of institutional-provided services including hospital coverage and a limited nursing home benefit provided to beneficiaries at a zero premium), has always been a voluntary choice. Part B enrollees pay a premium for such coverage. Given that the US government essentially subsidizes 75% of Part B outlays, the financial terms of Part B coverage are generally viewed as attractive, and 95% of Medicare-eligible persons enroll in it. Individuals who delay enrollment may face a 10% increase in Medicare Part B premiums for each 12-month period that they could have had Part B but did not enroll.² An exception is that this penalty does not apply if the individual has insurance coverage through his or her own or spouse’s current employer. For most full-time employees in large firms, however, Medicare benefits tend to be less generous than benefits provided by health insurance plans available through their employers (Kaiser Family Foundation Issue Brief No. 7768-02). For most working individuals who are also age-eligible for Medicare, employer-sponsored health insurance (ESHI) would be their primary payer and Medicare Part B would be the secondary payer. Large firms, defined as those with more than 20 employees, are required by law to offer Medicare-eligible employees the same benefits as younger employees. Since this law does not apply to small firms, however, Medicare is likely to be the primary payer for individuals working at such firms. By contrast, for retirees who are provided supplemental insurance coverage by their former employer, Medicare is typically the primary payer. The supplemental plans cover some services not covered by Medicare and some cost-sharing is imposed by Medicare on beneficiaries.

Higher Medicare premiums resulting from means testing may affect the decision to work and, conditional on working, the number of hours an individual chooses to work. An increase in premiums decreases future disposable income and thereby creates an incentive for some individuals to increase labor supply to accumulate enough wealth to live comfortably after retiring (income effect). Moreover, by reducing the price of ESHI relative to Medicare, means-tested premiums might induce some individuals to continue working in order to keep access to ESHI (substitution effect). The substitution effect does not apply to people not yet eligible for Medicare (those under age 65). Therefore, for this group of individuals, Medicare premiums affect labor supply solely via the income effect. In our empirical analysis we assess the impact of Medicare premiums separately for persons younger than age 65 and those 65 years or older. Both the income and substitution effects suggest that individuals would be more likely to postpone retirement (extensive margin) or work longer hours (intensive margin). Alternatively, some individuals may retire, reduce work hours, or switch to a lower-wage position (i.e., purchase non-wage benefits) in order to pay lower Part B premiums. Therefore, the net impact of increased premiums on labor supply is theoretically ambiguous.

Several previous studies have examined how the retirement decision or its timing is affected by the availability of post-retirement insurance. Studies have examined the impact of Medicare (Rust and Phelan 1997), employer-sponsored retiree coverage (Karoly and Rogowski 1994; Rogowski and Karoly 2000; Blau and Gilleskie 2001, 2008), expansion of Veterans’ Affairs coverage of both services and population covered (Boyle and Lahey 2010), and continuity of coverage mandates (Gruber and Madrian 1995). These studies tend to find that greater availability of post-retirement insurance coverage (or availability of more generous coverage) leads to earlier retirement. By contrast, the empirical literature on the relationship between other means-tested public health insurance programs, primarily Medicaid, and labor supply indicates mixed results (Winkler 1991; Yelowitz 1995; Meyer and Rosenbaum 2001; Borjas 2003; Strumpf 2011; Baicker, Finkelstein, Song, and Taubman 2014; Garthwaite, Gross, and Notowidigdo 2014).

Fewer studies have gauged the impact of changes in the price of insurance on LFP. Johnson, Davidoff, and Perese (2003) examined the impact of higher post-retirement health insurance premiums relative to pre-retirement/ESHI premiums for individuals who were not yet eligible for Medicare. They found that higher premiums significantly reduced the probability of retirement. Baicker and Chandra (2006) found that higher private health insurance premiums decreased the probability of being employed, decreased hours worked, and increased the probability of part-time work. To our knowledge, no existing studies have examined the impact of Medicare means-tested premiums on retirement decisions.

Empirical Methodology

Data

We use data from the 1992 to 2010 waves of the HRS, a nationally representative biennial survey of individuals 51 years and older and their spouses/partners who were of any age (Juster and Suzman 1995). We use data from the RAND HRS (version M) file, which consists of 36,986 individuals interviewed at least once from 1992 to 2010. We restrict our sample to individuals aged 55 to 80 and exclude those who never worked or were not US residents at the time of the interview. We also exclude persons with Medicaid coverage, since it pays Medicare premiums on behalf of beneficiaries. Observations with missing values on key variables, for example, marital status, retirement status, income and sociodemographic characteristics, are omitted. Finally, since our empirical methodology requires at least three consecutive waves of data on each sample person, we exclude those who did not meet this requirement. These exclusions reduce the analysis sample to 17,235 individuals and 68,010 person-year observations.

Measures of Labor Supply

Retirement status is captured by a binary variable set to 1 if the person is fully retired (i.e., not working, not looking for work, and a response that she or he is retired). To measure the intensive margin of labor supply, we use three variables for the sample of persons who report working. The first is a binary variable for part-time work, with full-time work forming the reference category. The second is the usual number of hours per week that the individual works at his or her main job, conditional on working at all. The third is the usual number of weeks per year at the person’s main job, also conditional on any work.

Explanatory Variables

The key explanatory variable is the monthly Medicare Part B premium that each individual expects to pay that year. Before 2007, all Medicare beneficiaries paid the same monthly premium. Starting in 2007, higher-income beneficiaries were assessed higher premiums depending on whether their modified adjusted gross incomes (MAGI) exceeded certain statutory thresholds, which differed for beneficiaries filing joint or individual tax returns. For example, in 2010, for those filing individually, the premium for beneficiaries with a MAGI less than or equal to $85,000 was $110.50/month, and $353.60 for those with a MAGI greater than $214,000. Table 1 presents the income-related premium structure in 2010.

Table 1.

Part B Premium Structure in 2010

Beneficiaries who file an individual tax return with income:	Beneficiaries who file a joint tax return with income:	Income-related monthly adjustment amount	Total monthly premium amount
≤ $85,000	≤ $170,000	$ 0.00	$110.50
> $85,000 and ≤ $107,000	> $170,000 and ≤ $214,000	$44.20	$154.70
> $107,000 and ≤ $160,000	> $214,000 and ≤ $320,000	$110.50	$221.00
> $160,000 and ≤ $214,000	> $320,000 and ≤ $428,000	$176.80	$287.30
> $214,000	> $428,000	$243.10	$353.60

Beneficiaries who are married but file a separate tax return from their spouse:	Income-related monthly adjustment amount	Total monthly premium amount
≤ $85,000	$ 0.00	$110.50
> $85,000 and ≤ $129,000	$176.80	$287.30
> $129,000	$243.10	$353.60

Sources: Centers for Medicaid and Medicare Services (CMS) press release, accessed at http://www.cms.gov/Newsroom/MediaReleaseDatabase/Fact-Sheets/2009-Fact-Sheets-Items/2009-10-16.html (September 24, 2012). Information on the premium structure in 2008 is available at https://www.cms.gov/Newsroom/MediaReleaseDatabase/Fact-sheets/2007-Fact-sheets-items/2007-10-01.html.

Since the HRS contains neither MAGI nor actual premiums paid, we impute the monthly Medicare Part B premium for each individual based on the person’s total household income, a proxy for MAGI, and marital status, a proxy for tax filing status. We discuss the implications of measurement error in the premium variable below. The premium is based on income reported on the most recent tax return, generally income received two years previously. For example, 2010 premiums are generally based on 2009 tax returns, which are based on 2008 income. Therefore, we use income and marital status from the preceding wave to calculate thresholds used to determine Part B premiums. Since the HRS is a biennial survey, a one-wave lag is equivalent to a two-year lag. The Appendix provides further details on the construction of the premium variable. The explanatory variable is the monthly premium that an individual can expect to be charged based on their income and marital status, if they enrolled in Medicare Part B (not actual premiums paid). Table 2 presents monthly premiums by year and demonstrates variation over time and across beneficiaries.

Table 2.

Monthly Medicare Part B Premiums (in Nominal $) by Year

Year	Low	Low-Medium	Medium	Medium-High	High
1992	31.80	31.80	31.80	31.80	31.80
1994	41.10	41.10	41.10	41.10	41.10
1996	42.50	42.50	42.50	42.50	42.50
1998	43.80	43.80	43.80	43.80	43.80
2000	45.50	45.50	45.50	45.50	45.50
2002	54.00	54.00	54.00	54.00	54.00
2003	Medicare Modernization Act enacted
2004	66.60	66.60	66.60	66.60	66.60
2006	88.50	88.50	88.50	88.50	88.50
2007	Income-related premiums effective
2008	96.40	122.20	160.90	199.70	238.40
2010	110.50	154.70	221.00	287.30	353.60

We estimate separate regressions for persons under and over age 65. All regressions include a basic set of sociodemographic characteristics—age, gender, race and ethnicity, and years of education. Since eligibility for Social Security benefits provides an additional incentive to retire at specific ages, we use categorical age variables to capture such incentives. Specifically, in the analysis of persons under age 65, we use an indicator for age 62, the earliest age at which one can begin receiving Social Security benefits, and an indicator for ages 63 to 64, with ages 55 to 61 forming the reference category. For the analysis of persons age 65 and older, we use separate indicators for ages 66, 67, 68 to 70, and over 70, with age 65 as the reference category. The full retirement age (FRA) at which a person becomes eligible for full Social Security benefits is 65 for persons born in 1937 or earlier. It increases by two months for each additional year of birth for those born after 1937, up to age 67. Individuals who delay retirement past their FRA receive larger benefits up to age 70; delaying receipt past age 70 provides no additional credit. The categorical age variables account for these varying retirement incentives by age created by Social Security eligibility rules. Finally, we include the monthly unemployment rate and housing price index in the person’s census division of residence to account for intertemporal and geographic variation in the impact of recessions. Recessions have been linked to LFP in general and to retirement in particular (Coile and Levine 2011a, 2011b; Goda, Shoven, and Slavov 2011; McFall 2011). The unemployment rate was obtained from the Bureau of Labor Statistics website, and the (purchase-only) housing price index was obtained from the Federal Housing Finance Agency website. These variables were then merged with HRS data using the month and year of interview and census division of residence. Since the public-use HRS data set does not provide information on finer geographic areas, such as state or county, we are limited in the geographic variation we can use.

Table 3 presents summary statistics for the full sample of persons and separately for those age 65 and over and those under age 65, respectively. Most individuals in the full sample are retired (55%), and a higher share of the older sample is retired (68%) compared to the younger sample (32%). The mean monthly Part B premium is $80 (2010 US dollars). In the full sample, 9.4% of individuals face higher Part B premiums post-2007, which is attributable to MMA policy changes. Among Medicare age-eligible persons, 7.7% face higher premiums. These rates are slightly higher than the 4% to 5% reported by other sources, likely because our sample excludes Medicaid beneficiaries who would not have been subject to these higher rates. A higher share of persons in the under-65 sample face higher premiums (14.3%), probably because mean income tends to be higher for persons in this age group.

Table 3.

Descriptive Statistics

	Full sample	Age ≥ 65	Age < 65
Retired	0.5461	0.6745	0.3168
Medicare Part B premium (in 2010 US dollars)	80.1471 (27.5287)	82.5052 (26.6981)	75.9345 (28.4718)
Fraction facing higher income-related premiums	0.0941	0.0772	0.1434
Age	67.9283 (6.1766)	71.5409 (4.6276)	61.4743 (1.7547)
Male	0.4669	0.4753	0.4520
Black	0.1178	0.1094	0.1327
Other race	0.0187	0.0159	0.0238
Hispanic	0.0646	0.0569	0.0783
Years of education	12.6501 (2.9546)	12.5548 (2.9655)	12.8202 (2.9274)
Unemployment rate	5.5478 (1.7557)	5.6585 (1.8017)	5.3500 (1.6521)
Observations	68,010	43,603	24,407
Persons	17,235	13,385	10,915

Notes: Standard deviations for non-binary variables are presented in parentheses.

Estimation

To evaluate the impact of Part B premiums on labor supply, we first estimate parameters of the labor supply function using ordinary least squares (OLS):

L S_{i t} = β_{0} + β_{1} P r e m i u m_{i t} + β_{2} X_{i t} + T_{t} + ε_{i t}

(1)

where $L S$ represents labor supply for person i in year t and $X$ a set of sociodemographic variables, the unemployment rate, and the housing price index. Year fixed effects account for secular trends in labor supply ( $T$ ), and $ε$ represents the error term. Since Equation (1) includes year fixed effects and premiums do not vary among beneficiaries before 2007, this specification exploits variation in premiums by MAGI and tax-filing status introduced by the MMA in 2003. Since the substitution effect only applies to beneficiaries ages 65 and over, we expect the impact of premiums on retirement to differ between persons under and over age 65.

Although Medicare premiums are statutorily determined, parameter estimates from Equation (1) may not identify causal effects. Premiums are based on income and tax-filing status, both of which are likely to be endogenous to labor supply decisions. Unobserved shocks to income would influence the premiums faced by individuals as well as retirement decisions. For example, the 2007−2009 recession may have differentially affected high-income versus medium- to low-income individuals in ways not fully captured by the year fixed effects or the unemployment rate and housing price index. High-income persons may have experienced larger declines in wealth during the Great Recession, causing them to delay retirement. However, they may still face high Part B premiums when they do retire, leading to a spurious correlation between such premiums and retirement. A second concern relates to measurement error in the premium variable. Household income and marital status are imperfect proxies for MAGI and tax-filing status and, therefore, some individuals may be misclassified as facing higher or lower premiums than they actually do. The direction and frequency of such misclassifications is unclear. To the extent that this measurement error is random, it introduces attenuation bias, which implies that $β_{1}$ is likely to be a lower bound of the true effect of premiums on retirement.

We address these concerns by using a simulated instrumental variables (IV) method based on the approach used by Dahl and Lochner (2012). To account for individual specific unobserved factors, we employ a first differences (FD) specification:

Δ L S_{i t} = β_{0} + β_{1} Δ P r e m i u m_{i t} + β_{2} Δ X_{i t} + β_{3} w_{i} + T_{t} + Δ ε_{i t}

(2)

Equation (2) includes time-invariant observable characteristics ( $w_{i}$ ), which affect both current labor supply and the change in labor supply. The FD specification represents an improvement over its OLS counterpart in accounting for time-invariant factors, unobservable to researchers, that may be correlated with both income and labor supply. For example, the FD specification accounts for systematically different preferences for leisure between high- and low-income persons. It does not account, however, for time-varying unobserved shocks to income or marital status that may also be correlated with labor supply. Therefore, we employ a simulated IV approach that exploits exogenous variation in the Part B premium structure resulting from the MMA. Part B premiums are a function of MAGI and tax-filing status and can be denoted as:

P r e m i u m_{i t} = f_{t}^{T S_{i t}} (M A G I_{i t})

where $M A G I_{i t}$ is modified adjusted gross income for person $i$ in year $t$ . The function $f_{t}^{T S_{i t}} (\cdot)$ represents the premium structure in year $t$ for a person with tax-filing status $T S_{i t}$ . Since premiums were the same for everyone prior to the MMA, $f_{t}^{T S_{i t}} (M A G I_{i t}) = c_{t}$ , for $t < 2007$ . The instrument is defined as:

Δ P_{i t} = f_{t}^{T S_{i t - 1}} (M A G I_{i t - 1}) - f_{t - 1}^{T S_{i t - 1}} (M A G I_{i t - 1})

This instrument represents the predicted increase in premiums, if MAGI and tax-filing status stay the same as they were in the prior year but the premium structure changes. Thus, $Δ P_{i t}$ captures changes in premiums faced by individuals resulting only from exogenous policy-induced changes in the premium structure, and not changes in MAGI or tax-filing status. In practice, we use income and marital status information acquired from interviews conducted two years previously, as proxies for $M A G I_{i t}$ and $T S_{i t}$ . Since the HRS is a biennial survey, we use $M A G I_{i t - 2}$ (instead of $M A G I_{i t - 1}$ ) to construct the IV and use income and marital status from four years previously for $M A G I_{i t - 2}$ and $T S_{i t - 2}$ . For example, MAGI in 2010 is measured by 2008 income and lagged MAGI in 2010 by 2006 income. Therefore, at least three waves of data are required for each sample person.

The simulated IV approach has been used to study the effects of income or income-related variables (e.g., Medicaid eligibility) by using exogenous policy changes in Medicaid eligibility (Currie and Gruber 1996), tax reform (Gruber and Saez 2002), and the Earned Income Tax Credit (Dahl and Lochner 2012). The key identifying assumption is that the instrument ( $Δ P_{i t}$ ) is not correlated with the error term $Δ ε_{i t}$ . That is, the policy-induced increase in premiums should not be correlated with any unobserved shocks that also affect retirement. Using the above example, any differential effects of a recession on high-income persons should not be correlated with the change in premiums they would face if their income stayed at the same level it was in the last period. Earlier research has pointed out that since the IV is based on lagged income, IV estimation of Equation (2) may lead to biased results if unobserved factors are correlated with lagged income and the error term (e.g., see Gruber and Saez 2002). Serially correlated income shocks, measurement error, and regression to the mean are potential sources of bias. For example, if persons with high income in 2006 were more affected by the Great Recession in 2008 (serially correlated income shocks), then they may be more likely to delay retirement relative to low-income persons. In such a case, the instrument based on income in 2006 would be correlated with the unobserved shock in 2008 leading to biased parameter estimates. Therefore, using the solution in prior studies, we add a flexible function of lagged income, $Φ (M A G I_{i t - 1})$ , to Equation (2) when employing the IV strategy. The flexible polynomial serves as a control function that captures the true relationship between labor supply shocks and lagged income. Thus, the identifying assumption of the simulated IV specification is that, conditional on $Φ (M A G I_{i t - 1})$ , changes in the premium structure are not correlated with idiosyncratic shocks affecting income and labor supply changes. That is, we assume that any unobserved differential trends in retirement by income are fully captured by the flexible polynomial. If 2008 recessionary shocks are correlated with 2006 income, then $Φ (M A G I_{i t - 1})$ fully accounts for such correlations.

We use a fourth-order polynomial in lagged income and indicators for lagged marital status as part of $Φ (M A G I_{i t - 1})$ .³ Since having a correct specification of $Φ (M A G I_{i t - 1})$ is crucial for satisfying the identification assumption, we also perform robustness checks in which we include interactions between terms in $Φ (M A G I_{i t - 1})$ and other covariates (e.g., age, gender, race). Further, we estimate specifications that assess whether differential effects of recessions could be biasing our results. Pfeffer, Danziger, and Schoeni (2013) found that higher-wealth households experienced larger absolute declines in wealth, but lower relative declines, during the Great Recession. Households with a greater share of assets invested in stocks or home equity would have experienced greater wealth shocks. Therefore, we estimate models that control for total wealth, the ratio of wealth in equities to total wealth, and the ratio of wealth in Individual Retirement Accounts (IRA)/Keogh accounts to total wealth. If $Φ (M A G I_{i t - 1})$ fully captures differential trends in retirement due to unobserved income shocks as in the previous example, then controlling for these additional variables should not significantly change our results.

Results

Table 4 presents OLS and FD results for the sample of individuals aged 65 and older. A $10 increase in monthly Medicare Part B premiums decreases the share of retired persons by 0.84 percentage points in the OLS specification (column (1)). When we use the FD specification (column (2)), the impact of Part B premiums becomes much smaller in magnitude and is no longer statistically significant. Dahl and Lochner (2012) noted that attenuation bias due to measurement error is likely to be exacerbated in the differenced model compared to the model in levels. This may explain the smaller effects with FD. In any case, both OLS and FD estimates are probably biased due to unobserved time-varying shocks, and the direction of this bias is unclear. Therefore, we turn next to the FD-IV estimates in Table 5.

Table 4.

Effect of Part B Premiums on Retirement for 65+ Sample (OLS and FD Estimates)

	(1)OLS	(2)FD
Dependent variable:	Retired	$Δ$ Retired
Part B premium (in hundreds, 2010 US dollars)	−0.0837*** (0.0144)
$Δ$ Part B premium (in hundreds, 2010 US dollars)		−0.0111(0.0125)
Age = 66	0.0390***
	(0.0110)
Age = 67	0.0503***
	(0.0086)
Age = 68 to 70	0.1049***
	(0.0088)
Age ≥ 71	0.2140***
	(0.0090)
$Δ$ Age = 66		0.0379***
		(0.0071)
$Δ$ Age = 67		0.0245***
		(0.0073)
$Δ$ Age = 68 to 70		0.0349***
		(0.0083)
$Δ$ Age ≥ 71		0.0359***
		(0.0116)
Male	0.0434***	0.0042
	(0.0069)	(0.0027)
Black	−0.0103	0.0028
	(0.0119)	(0.0046)
Other race	−0.0859***	−0.0233**
	(0.0320)	(0.0102)
Hispanic	−0.0541***	0.0146**
	(0.0153)	(0.0067)
Years of education	0.0029**	−0.0010*
	(0.0013)	(0.0005)
Unemployment rate	0.0106***
	(0.0041)
$Δ$ Unemployment rate		−0.0031
		(0.0033)
Housing Price Index	0.0002
	(0.0002)
$Δ$ Housing Price Index		−0.0003*
		(0.0001)
Constant	0.4672***	0.0828***
	(0.0510)	(0.0239)
R ²	0.0633	0.0057
N	43,603	43,603
Persons	13,385	13,385
Households	9,285	9,285

Notes: Robust standard errors in parentheses are clustered at the household level. Regressions also include year fixed effects, a fourth-order polynomial in lagged income, and lagged indicators for married and single (separated forms the reference group). FD, first differences; OLS, ordinary least squares.

p < 0.10; **p < 0.05; ***p < 0.01.

Table 5.

Impact of Medicare Part B Premiums on Retirement for 65+ Sample (FD-IV Estimates)

	(1)	(2)
Dependent variable:	$Δ$ Part B premium	$Δ$ Retired
	1st stage	2nd stage
$Δ$ Part B premium (in hundreds, 2010 US dollars)		−0.1629**(0.0830)

Instrument	0.0022***
	(0.0004)
$Δ$ Age = 66	0.0115***	0.0397***
	(0.0028)	(0.0072)
$Δ$ Age = 67	0.0085***	0.0260***
	(0.0028)	(0.0073)
$Δ$ Age = 68 to 70	0.0131***	0.0374***
	(0.0037)	(0.0084)
$Δ$ Age ≥ 71	0.0108**	0.0379***
	(0.0051)	(0.0117)
Male	0.0023	0.0045
	(0.0012)	(0.0027)
Black	−0.0039**	0.0023
	(0.0019)	(0.0046)
Other race	0.0086	−0.0220**
	(0.0070)	(0.0103)
Hispanic	−0.0009	0.0146**
	(0.0019)	(0.0068)
Years of education	0.0023***	−0.0007
	(0.0003)	(0.0005)
$Δ$ Unemployment rate	−0.0011	−0.0033
	(0.0022)	(0.0034)
$Δ$ Housing Price Index	−0.0002**	−0.0003**
	(0.0001)	(0.0002)
Constant	−0.0438***	0.0774***
	(0.0108)	(0.0241)
Kleibergen-Paap F-statistic	29.9875
N	43,603
Persons	13,385
Households	9,285

p < 0.10; **p < 0.05; ***p < 0.01.

For the sample of persons aged 65 years or older, we find a significant negative impact of Part B premiums on retirement when using the simulated IV model to account for the endogeneity of premiums. A $10 increase in monthly premiums reduces retirement rates by 1.63 percentage points (Table 5, column (2)). This result is consistent with the hypothesis that as the price of Medicare increases relative to employer-sponsored insurance, older workers increase labor supply to retain access to this insurance. To assess the magnitude of the point estimate, we calculate the semi-elasticity of labor supply. Relative to a mean Part B premium of $82.50 for this sample, our parameter estimate implies that a 10% increase in Medicare premiums reduces retirement rates by 1.32 percentage points. These results are consistent with findings of earlier studies showing that labor supply is sensitive to changes in insurance premiums. For example, Baicker and Chandra (2006) found that a 10% increase in private health insurance premiums reduced the probability of employment by 1.2 percentage points. One concern with this comparison is that the researchers examined the labor supply of persons aged 22 to 64, who might have had very different labor supply responses compared to the elderly. Therefore, we also calculate the income elasticity of labor supply and compare our results to estimates from the literature on retiree health insurance. The average wage income for our sample is $6,255 (2010 US dollars). Thus, a $10 increase in monthly premiums (or a $120 increase in annual premiums) represents 1.9% of annual wage income. From Table 3 we see that approximately 67% of the elderly are retired. Therefore, a decline of 1.63 percentage points in retirement represents a 2.4% decline in retirement rates. This implies that the LFP income elasticity for persons aged 65 and over is −1.26. This estimate is slightly larger than the −0.8 elasticity identified by Fitzpatrick (2014), who evaluated the impact of retiree health insurance on LFP of public school employees aged 50 years or older. Given the large standard errors, a −0.8 elasticity falls within the 95% confidence interval of our estimate. However, it is also plausible that the labor supply of persons aged 65 and older is more price-sensitive than that of the slightly younger population examined by Fitzpatrick (2014).

Results based on the IV specification indicate a much larger effect of premiums compared to their OLS or FD counterparts. This finding is consistent with the hypothesis that the OLS and FD results suffer from attenuation bias due to measurement errors in premiums. Unobserved factors may also bias the OLS/FD estimates downward. Further, the IV specification yields a Local Average Treatment Effect (Imbens and Angrist 1994). It is plausible that the impact of premiums is much larger for persons directly affected by the policy change (i.e., the “compliers”). Higher-income individuals may be more responsive to increases in Medicare premiums because they tend to be healthier. Therefore, they can more easily increase their labor supply and/or have more opportunities for obtaining employer-sponsored health insurance.

Next, we assess the validity of our IV strategy. The instrument based on lagged income is strongly correlated with the change in premiums in the first stage (Table 5, column (1)), and the Kleibergen-Paap F-statistic is 29.9875, above the commonly accepted threshold of 10 and also above the Stock-Yogo critical values (Stock and Yogo 2002).

The analysis so far simply captures a change in retirement status; it combines cases in which persons who are working delay retirement in response to higher premiums with cases in which individuals who have already retired return to work in response to higher premiums. To distinguish between these two channels, we divide the sample by retirement status at the previous interview and estimate the FD-IV model separately for each subsample. The model presented in column (1) of Table 6 restricts the sample to persons who were not (fully) retired at the time of the previous survey, thus capturing delays in full retirement. The model in column (2) presents results for the sample of individuals who were fully retired at the time of the previous survey, capturing re-entry into the workforce by retirees. We find a marginally significant effect (at the 10% level) of premiums for the sample of non-retirees, whereas in the case of retirees the coefficient on the premium variable is small in magnitude and not statistically significant. This finding suggests that the main channel through which Medicare Part B premiums affect labor supply is through delays in retirement for persons who are not yet retired and not via retirees returning to work.

Table 6.

Evaluating Delays in Retirement versus Return to Work for 65+ Sample

	(1)	(2)
	Delay retirement	Return to work
$Δ$ Part B premium (in hundreds, 2010 US dollars)	−0.2059*(0.1152)	−0.0271(0.2275)

$Δ$ Age = 66	−0.0572***	−0.0162
	(0.0174)	(0.0114)
$Δ$ Age = 67	−0.0639***	−0.0354***
	(0.0128)	(0.0085)
$Δ$ Age = 68 to 70	−0.0945***	−0.0601***
	(0.0192)	(0.0108)
$Δ$ Age ≥ 71	−0.1133***	−0.0745***
	(0.0263)	(0.0132)
Male	−0.0466***	0.0528***
	(0.0098)	(0.0044)
Black	0.0104	0.0067
	(0.0166)	(0.0071)
Other race	−0.0554*	−0.0206
	(0.0322)	(0.0217)
Hispanic	0.0143	−0.0359***
	(0.0213)	(0.0115)
Years of education	−0.0022	0.0038***
	(0.0019)	(0.0009)
$Δ$ Unemployment rate	−0.0048	0.0030
	(0.0068)	(0.0031)
$Δ$ Housing Price Index	−0.0002	−0.0002
	(0.0003)	(0.0001)
Constant	0.4394***	−0.1825***
	(0.0532)	(0.0319)
Kleibergen-Paap F-statistic	24.2720	2.8684
N	10,629	19,438
Persons	4,940	7,914
Households	4,132	6,083

p < 0.10; **p < 0.05; ***p < 0.01.

In Table 7, we present evidence on the impact of premiums on the intensive margin of labor supply. Overall, the estimates are consistent with the hypothesis that higher premiums increase labor supply. Higher premiums reduce the probability of working part-time, conditional on working, and lead to increases in the hours of work per week, and in the number of weeks worked per year. None of these estimates, however, are statistically significant at conventional levels.⁴ Conditioning on working significantly reduces the sample size, and lack of adequate statistical power may explain the imprecise estimates.

Table 7.

Impact of Premiums on the Intensive Margin of Labor Supply for 65+ Sample (FD-IV Estimates)

	(1)	(2)	(3)
Dependent variable:	$Δ$ Work part time	$Δ$ Hours per week	$Δ$ Weeks per year
$Δ$ Part B premium (in hundreds, 2010 US dollars)	−0.1552(0.1011)	4.7209(3.3795)	0.1830(2.5323)
$Δ$ Age = 66	0.0317**	−0.7468*	0.3242
	(0.0126)	(0.3954)	(0.2956)
$Δ$ Age = 67	0.0212	−0.5157	−0.0933
	(0.0144)	(0.4554)	(0.3630)
$Δ$ Age = 68 to 70	0.0279	−0.8728	0.3777
	(0.0174)	(0.5394)	(0.4290)
$Δ$ Age ≥ 71	0.0106	−1.0626	0.5675
	(0.0260)	(0.8145)	(0.7123)
Male	0.0111	−0.5090**	0.1210
	(0.0069)	(0.2264)	(0.1973)
Black	−0.0367***	1.2740***	0.3269
	(0.0103)	(0.3170)	(0.3013)
Other race	0.0101	0.1368	0.2951
	(0.0244)	(0.5904)	(0.5435)
Hispanic	0.0226	−0.6577	0.2504
	(0.0169)	(0.4689)	(0.4311)
Years of education	0.0019	−0.0743*	−0.0679*
	(0.0013)	(0.0443)	(0.0358)
$Δ$ Unemployment rate	0.0052	0.0512	−0.0080
	(0.0072)	(0.2191)	(0.1841)
$Δ$ Housing Price Index	−0.0000	−0.0183*	0.0130
	(0.0003)	(0.0106)	(0.0091)
Constant	0.0405	−1.5231	0.3462
	(0.0512)	(1.6051)	(1.1636)
Kleibergen-Paap F-statistic	33.1864	39.4408	30.0054
N	9,299	8,742	8,799
Persons	3,938	3,803	3,802
Households	3,421	3,310	3,310

p < 0.10; **p < 0.05; ***p < 0.01.

We also examine the impact of Medicare premiums on persons who are younger than age 65 and do not have Medicare coverage due to disability or other reasons (Table 8). These individuals are not yet, but soon will be, eligible for Medicare. Therefore, the impact of higher premiums for this sample is likely to operate solely through the income effect, as the substitution effect is not yet relevant for them. Individuals who would have opted for early retirement in the absence of means-tested premiums might decide to work longer to be able to afford the higher premiums they must pay after age 65. Limiting the sample to persons under age 65 results in a negative effect of premiums on the probability of retiring early in the OLS specification. This effect, however, becomes statistically insignificant when we account for the endogeneity of premiums. To the extent that individuals are forward-looking, these results, together with the findings in Table 5, suggest that the substitution effect is likely to be more important for labor supply decisions than the income effect. However, it is also plausible that Medicare premiums are not particularly salient in the early retirement decisions of persons younger than age 65, or that the magnitude of the income change represented by higher Part B premiums is not large enough to influence their retirement decisions.

Table 8.

Effect of Part B Premiums on Retirement for Individuals Aged < 65

	(1)	(2)	(3)
Dependent variable:	Retired	$Δ$ Retired	$Δ$ Retired
	OLS	FD	FD-IV
Part B premium (in hundreds, 2010 US dollars)	−0.0402**(0.0177)
$Δ$ Part B premium (in hundreds, 2010 US dollars)		−0.0079(0.0153)	0.1000(0.1050)

Age = 62	0.1443***
	(0.0081)
Age ≥ 63	0.2094***
	(0.0062)
$Δ$ Age = 62		0.0777***	0.0777***
		(0.0071)	(0.0071)
$Δ$ Age ≥ 63		0.0932***	0.0934***
		(0.0080)	(0.0080)
Male	−0.0276***	0.0090**	0.0085*
	(0.0072)	(0.0043)	(0.0044)
Black	−0.0174	0.0103	0.0111
	(0.0117)	(0.0070)	(0.0070)
Other race	−0.0477**	−0.0094	−0.0095
	(0.0241)	(0.0135)	(0.0136)
Hispanic	−0.0782***	−0.0101	−0.0098
	(0.0147)	(0.0090)	(0.0090)
Years of education	0.0023	−0.0024***	−0.0026***
	(0.0015)	(0.0009)	(0.0009)
Unemployment rate	0.0153***
	(0.0050)
$Δ$ Unemployment rate		−0.0006	−0.0011
		(0.0046)	(0.0047)
Housing Price Index	0.0001	−0.0006
	(0.0003)
$Δ$ Housing Price Index		−0.0001	−0.0001
		(0.0002)	(0.0002)
Constant	0.0988*	0.0814***	0.0836***
	(0.0566)	(0.0238)	(0.0240)
Kleibergen-Paap F-statistic			19.5330
N	22,291	22,291	22,291
Persons	10,122	10,122	10,122
Households	7,250	7,250	7,250

p < 0.10; **p < 0.05; ***p < 0.01.

Robustness Checks

In Table 9, we assess the robustness of our estimates to various specifications. An important assumption in our econometric strategy is that $Φ (\cdot)$ is correctly specified. Therefore, we estimate models adding interactions between each term in $Φ (\cdot)$ and age, indicators for gender, race and ethnicity, and years of education completed. Although the point estimate is slightly smaller in magnitude and less precisely estimated for some of the specifications, the overall conclusion that higher Medicare premiums reduce retirement rates is unchanged.⁵ In column (5) of Table 9, we examine the role of other concurrent policy changes that might affect labor supply. Specifically, the FRA has been increasing for individuals from recent birth cohorts reaching retirement age, and has been shown to affect the age at which individuals claim Social Security benefits (Behaghel and Blau 2012). The discrete age categories used in the regressions may not fully capture the different age and cohort incentives. Therefore, we estimate a specification in which we add indicators for each year of birth after 1937 (being born in 1937 or earlier forms the reference category). The results are robust to including birth-year dummies and, in fact, the estimated effect is slightly larger in magnitude.

Table 9.

Robustness Checks for 65+ Sample (FD-IV Estimates)

	(1)	(2)	(3)	(4)	(5)
Dependent variable:	$Δ$ Retired	$Δ$ Retired	$Δ$ Retired	$Δ$ Retired	$Δ$ Retired
$Δ$ Part B premium (in hundreds, 2010 US dollars)	−0.1675**(0.0844)	−0.1535*(0.0841)	−0.1501*(0.0833)	−0.1375(0.0850)	−0.2328**(0.0967)
$Φ (\cdot)$ × Age	X	X	X	X
$Φ (\cdot)$ × Male		X	X	X
$Φ (\cdot)$ × Race/Ethnicity			X	X
$Φ (\cdot)$ × Education				X
Birth-year dummies					X
Kleibergen-Paap F-statistic	29.8601	30.2375	31.9012	30.3394	24.2697
N	43,603	43,603	43,603	43,603	43,603
Persons	13,385	13,385	13,385	13,385	13,385
Household	9,285	9,285	9,285	9,285	9,285

Notes: Robust standard errors in parentheses are clustered at the household level. FD-IV, first difference instrumental variables.

p < 0.10; **p < 0.05; ***p < 0.01.

In Table 10, we further explore the robustness of our results to differential effects of the Great Recession. As mentioned earlier, a potential threat to identification is that the recession may have had a larger effect on persons facing high premiums. If such differential effects are not fully captured by the flexible polynomial in lagged income, unemployment rate, and housing price index variables, then the estimates in Table 5 would be biased. To address this concern, we draw on detailed measures of wealth and assets in the HRS. In column (1) of Table 10, we add the inflation-adjusted net value of total wealth (excluding a secondary residence) to our main specification. This variable is the sum of all wealth (housing, stocks, savings, and so forth) less all debt. In column (2), we add the ratio of wealth held in equity, defined as the net value of all stocks, mutual funds, and investment trusts, divided by the net value of total wealth. Column (3) adds the IRA ratio, which is the net value of all IRAs or Keogh accounts divided by the net value of total wealth. Finally, columns (4) through (7) add interactions between these asset variables and a continuous variable for year. These specifications test whether persons with riskier investments, who may have experienced larger shocks due to the recession, exhibit differential retirement trends compared to those with less-risky investments. Overall, the parameter estimates are robust across these alternative specifications; they are less precisely estimated, however, as we add more covariates.

Table 10.

Checking for Differential Recessionary Effects for 65+ Sample (FD-IV Estimates)

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Dependent variable:	$Δ$ Retired	$Δ$ Retired	$Δ$ Retired	$Δ$ Retired	$Δ$ Retired	$Δ$ Retired	$Δ$ Retired
$Δ$ Part B premium (in hundreds, 2010 US dollars)	−0.1667**(0.0820)	−0.1660**(0.0822)	−0.1656**(0.0822)	−0.2068*(0.1147)	−0.1598**(0.0843)	−0.1533*(0.0829)	−0.2001*(0.1146)
Total wealth	X	X	X	X			X
Stock ratio		X	X		X		X
IRA ratio			X			X	X
Total wealth × Year				X			X
Stock ratio × Year					X		X
IRA ratio × Year						X	X
Kleibergen-Paap F-statistic	31.0984	30.8435	30.8159	18.0678	28.9930	29.5863	17.8404
N	43,603	43,164	43,164	43,603	43,164	43,164	43,164
Persons	13,385	13,289	13,289	13,385	13,289	13,289	13,289
Households	9,285	9,203	9,203	9,285	9,203	9,203	9,203

Notes: Robust standard errors in parentheses are clustered at the household level. IRA, Individual Retirement Account; FD-IV, first difference instrumental variables.

p < 0.10; **p < 0.05; ***p < 0.01.

Heterogeneous Effects

Finally, we examine whether certain subpopulations are more sensitive to changes in Medicare premiums than others are (Table 11). Given the large literature on gender differences in labor supply, we first examine the impact of Medicare premiums separately for men and women. We find that women are more likely than men to delay retirement in response to premium increases. We do not find statistically significant effects, however, when we split the sample by gender. Next, we examine the role of access to other sources of insurance. Persons who have access to insurance through their spouses should exhibit less of a labor supply response to changes to the Medicare program compared to those who will be dependent on Medicare for insurance after retirement. Consistent with this hypothesis, we find a small and insignificant effect of Part B premiums on individuals who have employer-sponsored health insurance through their spouses. By contrast, those not covered by their spouses’ insurance significantly reduce retirement rates in response to premium increases.⁶ We also examine the role of (self-rated) health. The labor supply of elderly persons in good health, compared to less-healthy individuals, should be more responsive to increases in Medicare premiums, since they are more capable of increasing their work in response to premium changes. Individuals in poor health may have more difficulty dealing with job demands and therefore may not be able to adjust their labor supply easily. Consistent with this hypothesis, we find a large impact of premiums for persons who report excellent, very good, or good health. For those who report fair or poor health, the impact of premiums is small and insignificant. Finally, we examine differences by education. We find a marginally significant effect (at the 10% level) for persons with more than 12 years of education, and a small insignificant effect for those with 12 or fewer years of education. This finding suggests that individuals of lower socioeconomic status are less responsive to premium increases.

Table 11.

Heterogeneous Effects for 65+ Sample (FD-IV Estimates)

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Dependent variable:	$Δ$ Retired	$Δ$ Retired	$Δ$ Retired	$Δ$ Retired	$Δ$ Retired	$Δ$ Retired	$Δ$ Retired	$Δ$ Retired
Subsample:	Male	Female	Covered by spouse’s plan	Not covered by spouse’s plan	Excellent to good health	Fair to poor health	Years of education >12	Years of education ≤12
$Δ$ Part B premium (in hundreds, 2010 US dollars)	−0.1002	−0.2533	0.0018	−0.1994**	−0.1705*	−0.0379	−0.1493*	0.0338
$Δ$ Part B premium (in hundreds, 2010 US dollars)	(0.0780)	(0.2064)	(0.1716)	(0.0951)	(0.0970)	(0.1675)	(0.0853)	(0.4652)
Kleibergen-Paap F-statistic	39.2542	8.8507	13.8303	24.8179	23.6224	9.9351	28.5406	1.0883
N	20,723	22,880	5,760	37,456	32,419	11,156	17,820	25,783
Persons	6,312	7,073	2,514	12,380	11,004	5,617	5,339	8,046
Households	6,278	7,046	2,405	8,949	7,980	4,780	4,217	6,242

Notes: Robust standard errors in parentheses are clustered at the household level. FD-IV, first difference instrumental variables.

p < 0.10; **p < 0.05; ***p < 0.01.

Conclusion

We find that the means testing of Medicare Part B premiums delays retirement among older adults. We do not find significant effects of premiums on the intensive margin of labor supply, as measured by part-time work, weeks worked per year, or hours of work per week. We also find no effect of Medicare premiums for persons younger than age 65. Overall, our findings suggest that the elderly are considerably sensitive to changes in health insurance premiums, which is consistent with previous literature. For example, Buchmueller, Grazier, Hirth, and Okeke (2013) found that a $10 increase in monthly premium contributions led to a 2 to 3 percentage point decrease in the market share of retiree insurance plans. The main estimates, however, mask considerable heterogeneity in impact. Healthier individuals, persons without spousal coverage, and those with more than 12 years of education are more responsive to premium increases. This heterogeneity suggests that changes in Part B premiums may not result in such large labor supply responses among other populations. An important limitation of the HRS is that the sample is too small to compare differences by age, income, or other individual characteristics. Future research should consider larger data sets that may allow for such comparisons.

Our results have important implications for public policy because an increasing number of the elderly are expected to face higher premiums in the near future due to means testing. To the extent that lower retirement rates are associated with lower enrollment in Medicare Part B by high-income individuals, our findings suggest that the anticipated increase in revenue from means-tested premiums may not be realized. On the one hand, since high-income individuals also tend to be healthier, lower enrollment among this group would lead to increased adverse selection in the Medicare program. Therefore, means testing may lead to further financial pressure on Medicare, instead of alleviating the problem. On the other hand, longer work lives imply that these individuals will continue to pay payroll taxes, and the additional revenue from this may more than offset any loss in revenue from lower enrollment in Medicare Part B. In addition, Medicare expenditures will be lower for persons who continue working and have employer-sponsored coverage since Medicare is the secondary payer for these individuals. Further, as noted earlier, heterogeneous effects by socioeconomic status suggest that the labor supply effect of premium increases may be smaller if means testing is expanded to lower-income populations.

Longer work lives also confer other benefits in an aging society such as the United States. Lower retirement rates or longer work lives have implications for the Social Security program, which is also under considerable financial pressure. Delays in retirement may also affect individual health. If higher Part B premiums cause individuals to keep working and hence retain more generous ESHI plans, then this might lead to an improvement in health. In addition, being engaged in work has been shown to provide substantial benefits in terms of neurocognitive health (Rohwedder and Willis 2010). Our findings also have implications for labor markets. Since higher-income individuals are also likely to be more productive (assuming wages reflect marginal productivity), longer working lives for this group may lead to more productive labor markets. In summary, delayed retirement due to means-tested premiums may be beneficial but further research is required to evaluate the exact impact on public programs, population health, and labor markets.

Footnotes

Appendix

Acknowledgements

We are grateful for helpful comments and suggestions from Laura Dague, David Frisvold, conference participants at APPAM and ASHEcon, and seminar participants at the University of Iowa.

For information regarding the data and/or computer programs used for this study, please address correspondence to the authors at padmajaa@usf.edu.

1

CMS Media Release, Fact Sheet on “Medicare Premiums and Deductibles for 2012,” accessed at (January 14, 2013).

2

Accessed at (November 18, 2013).

3

Since the thresholds used to determine Part B premiums are based on nominal values of income, the variables in $Φ (M A G I_{i t - 1})$ are in nominal terms. In analysis not shown, we estimated models in which we used real income to construct $Φ (M A G I_{i t - 1})$ . The results were robust to adjusting for inflation.

4

A concern may be that older workers increase labor supply along the intensive margin in response to wage decreases driven by a change in labor demand. To address this concern, we also estimate regressions controlling for real wage. The estimates (available on request) are robust to including real wage.

5

In analysis not shown, we also estimated a model that included interactions between a linear time trend and $Φ (\cdot)$ . Because of a collinearity, we could only use a third-order polynomial in lagged income and indicators of lagged marital status as the control function. The effect of Medicare Part B premiums was –0.1216 but was not statistically significant (p value = 0.227).

6

Note that this group may include some individuals who have access to ESHI through their spouse but chose not to take it up.

References

Baicker

Katherine

Chandra

Amitabh

. 2006. The labor market effects of rising health insurance premiums. Journal of Labor Economics 24(3): 609–34.

Baicker

Katherine

Finkelstein

Amy

Song

Jae

Taubman

Sarah

. 2014. The impact of Medicaid on labor market activity and program participation: Evidence from the Oregon Health Insurance Experiment. American Economic Review 104(5): 322–28.

Behaghel

Luc

Blau

David M

. 2012. Framing Social Security reform: Behavioral responses to changes in the full retirement age. American Economic Journal: Economic Policy 4(4): 41–67.

Blau

David M.

Gilleskie

Donna B

. 2001. Retiree health insurance and the labor force behavior of older men in the 1990s. Review of Economics and Statistics 83(1): 64–80.

Blau

David M.

Gilleskie

Donna B

. 2008. The role of retiree health insurance in the employment behavior of older men. International Economic Review 49(2): 475–514.

Borjas

George J.

2003. Welfare reform, labor supply, and health insurance in the immigrant population. Journal of Health Economics 22: 933–58.

Boyle

Melissa A.

Lahey

Joanna N

. 2010. Health insurance and the labor supply decisions of older workers: Evidence from a U.S. Department of Veterans Affairs expansion. Journal of Public Economics 94(7–8): 467–78.

Buchmueller

Thomas C.

Grazier

Kyle

Hirth

Richard A.

Okeke

Edward N

. 2013. The price sensitivity of Medicare beneficiaries: A regression discontinuity approach. Health Economics 22(1): 35–51.

Coile

Courtney C.

Levine

Philip B

. 2011a. The market crash and mass layoffs: How the current economic crisis may affect retirement. BE Journal of Economic Analysis & Policy 11(1). doi: https://doi.org/10.2202/1935-1682.2568.

10.

Coile

Courtney C.

Levine

Philip B

. 2011b. Recessions, retirement, and social security. American Economic Review 101(3): 23–28.

11.

Currie

Janet

Gruber

Jonathan

. 1996. Saving babies: The efficacy and cost of recent changes in the Medicaid eligibility of pregnant women. Journal of Political Economy 104(6): 1263–96.

12.

Dahl

Gordon B.

Lochner

Lance

. 2012. The impact of family income on child achievement: Evidence from the Earned Income Tax Credit. American Economic Review 102(5): 1927–56.

13.

Fitzpatrick

Maria D.

2014. Retiree health insurance for public school employees: Does it affect retirement? Journal of Health Economics 38: 88–98.

14.

Garthwaite

Craig

Gross

Tal

Notowidigdo

Matthew J

. 2014. Public health insurance, labor supply, and employment lock. Quarterly Journal of Economics 129(2): 653–96.

15.

Goda

Gopi S.

Shoven

John B.

Slavov

Sita N

. 2011. What explains changes in retirement plans during the Great Recession? American Economic Review 101(3): 29–34.

16.

Gruber

Jonathan

Madrian

Brigitte C

. 1995. Health-insurance availability and the retirement decision. American Economic Review 85(4): 938–48.

17.

Gruber

Jonathan

Madrian

Brigitte C

. 2002. Health insurance, labor supply, and job mobility: A critical review of the literature. NBER Working Paper No. 8817. Cambridge, MA: National Bureau of Economic Research.

18.

Gruber

Jonathan

Saez

Emmanuel

. 2002. The elasticity of taxable income: Evidence and implications. Journal of Public Economics 84(1): 1–32.

19.

Imbens

Guido W.

Angrist

Joshua D

. 1994. Identification and estimation of local average treatment effects. Econometrica 62(2): 467–75.

20.

Johnson

Richard W.

Davidoff

Amy J.

Perese

Kevin

. 2003. Health insurance costs and early retirement decisions. Industrial and Labor Relations Review 56(4): 716–29.

21.

Juster

F. Thomas

Suzman

Richard

. 1995. An overview of the Health and Retirement Study. Journal of Human Resources 30(suppl.): 7–56.

22.

Kaiser Family Foundation Issue Brief No. 8126. 2010. Income-relating Medicare Part B and Part B premiums: How many Medicare beneficiaries will be affected?December. Accessed at https://www.kff.org/wp-content/uploads/2013/01/8126.pdf.

23.

Kaiser Family Foundation Issue Brief 7768-02. 2012. How does the benefit value of Medicare compare to the benefit value of typical large employer plans? A 2012 update. April. Accessed at https://www.kff.org/wp-content/uploads/2013/01/7768-02.pdf.

24.

Kaplan

Richard L.

2006. Means-testing Medicare: Retiree pain for little governmental gain. University of Illinois Legal Working Paper Series, 61. University of Illinois, Urbana-Champaign.

25.

Karoly

Lynn A.

Rogowski

Jeannette A

. 1994. The effect of access to post-retirement health insurance on the decision to retire early. Industrial and Labor Relations Review 48(1): 103–23.

26.

Madrian

Brigitte C.

1994. Employment-based health insurance and job mobility: Is there evidence of job-lock? Quarterly Journal of Economics 109(1): 27–54.

27.

McFall

Brooke H.

2011. Crash and wait? The impact of the Great Recession on the retirement plans of older Americans. American Economic Review 101(3): 40–44.

28.

Meyer

Bruce D.

Rosenbaum

Dan T

. 2001. Welfare, the Earned Income Tax Credit, and the labor supply of single mothers. Quarterly Journal of Economics 116(3): 1063–14.

29.

Penner

Rudolph G.

2011. Medicare premiums and Social Security’s cost-of-living adjustments. Program on Retirement Policy Research Report. Washington, DC: Urban Institute.

30.

Pfeffer

Fabian T.

Danziger

Sheldon

Schoeni

Robert F

. 2013. Wealth disparities before and after the Great Recession. ANNALS of the American Academy of Political and Social Science 650(1): 98–123.

31.

Rogowski

Jeannette

Karoly

Lynn

. 2000. Health insurance and retirement behavior: Evidence from the Health and Retirement Survey. Journal of Health Economics 19(4): 529–39.

32.

Rohwedder

Susann

Willis

Robert J

. 2010. Mental retirement. Journal of Economic Perspectives 24(1): 119–38.

33.

Rust

John

Phelan

Christopher

. 1997. How Social Security and Medicare affect retirement behavior in a world of incomplete markets. Econometrica 65(4): 781–831.

34.

Stock

James

Yogo

Motohiro

. 2002. Testing for weak instruments in linear IV regression. NBER Technical Working Paper No. 284. Cambridge, MA: National Bureau of Economic Research.

35.

Strumpf

Erin

. 2011. Medicaid’s effect on single women’s labor supply: Evidence from the introduction of Medicaid. Journal of Health Economics 30(3): 531–48.

36.

Winkler

Anne E.

1991. The incentive effects of Medicaid on women’s labor supply. Journal of Human Resources 26(2): 308–37.

37.

Yelowitz

Aaron S.

1995. The Medicaid notch, labor supply, and welfare participation: Evidence from eligibility expansions. Quarterly Journal of Economics 110(4): 909–39.